It's Actually Too Early to See Widespread AI Productivity Gains

“Today, you don’t see AI in the employment data, productivity data or inflation data,” says Torsten Slok, Apollo chief economist. “Similarly, for the S&P 493, there are no signs of AI in profit margins or earnings expectations.”

That is not without precedent, as that lag in quantifiable productivity impact also happened when computing technology was applied at work. In fact, it often happens that productivity actually decelerates when a new computing or other general-purpose technology is introduced.

GPTs are "consequential" innovations that transform entire economies over time.


source: MIT

So that J curve is not unusual.

But it might also be the case that productivity measurements are outdated. “It’s possible that “current measures of productivity do not capture the increases in value added that these technologies promote,” the McKinsey co-authors state. “Many new benefits are incorporated into products or services free of charge, for example, which means productivity statistics do not capture them.”

The best available evidence suggests that mismeasurement might explain up to 10 percent of the overall slowdown in productivity growth, a relevant but comparatively small effect,” they say.

Here’s a look at expected productivity gains from artificial intelligence. The impact might be less than you would expect.

source: Apollo Academy

Looking only at generative AI, there are clear and significant time savings, for example.



source: Visual Capitalist



source: Visual Capitalist

But those gains do not translate linearly into firm productivity statistics. Among the reasons: the need to recraft whole business processes (requiring new skills, organizational structure changes).

“General purpose technologies (GPTs) such as AI enable and require significant complementary

investments, including co-invention of new processes, products, business models and human

Capital,” say the authors of a paper published by the National Bureau of Economic Research. “These complementary investments are often intangible and poorly measured in the national accounts, even when they create valuable assets for the firm.”

Also, keep in mind that “whole economy” productivity tends to improve at rates between one percent and two percent annually, over time.

source: St. Louis Federal Reserve

So some economists note that measurable or quantifiable gains from other earlier GPTs took decades, though the impact of computing technologies happened much faster.

source: JP Morgan.

As noted above, AI impact on tasks can be quite high, but the impact on gross national product or productivity will not track in linear fashion. Greater output with similar or less input leads to measurable productivity gains only when the output affects sales and other revenue-related activity.

Isolating the impact of particular inputs requires us to make judgments. When multiple processes change, how do we evaluate the individual impact? If sales channels, production processes, marketing, advertising, applied AI, headcount and customer demand all change at once, any estimation of input factor contribution is subjective.

But in any case, it actually is too early to document AI-driven productivity increases. And the actual impact could well be negative.

Why the Walk for Peace Might Have Touched People

Many of us arguably have been pleasantly surprised by the emotional and apparently widespread reaction to the Walk for Peace: 20 monks and Aloka the “peace dog” on a 2,300-mile walk "to promote national healing, unity and compassion."

“The message of peace and mutual understanding conveyed through their conduct, marked by humility and calm presence, has resonated with many people they encountered along the route,” noted Tencho Gyatso, a niece of the Dalai Lama. 

But why?

Perhaps because the walk reflects:

• Spiritual hunger

• Fatigue with ideological conflict

• Desire for authenticity

• Attraction to embodied sacrifice

• Longing for visible goodness.

 

source 

Observers of culture and religion might say a variety of possible reasons contributed. The symbolism of unity; sacrifice; the moral beauty; sacred moments that rupture everyday space; the sense of pilgrimage; unmet spiritual hunger; subconscious response to sages, ascetics and monks;  .    

“When a group of monks walks quietly, consistently, without obvious self-promotion, that creates what sociologists call moral coherence.” 

People sense integrity and authenticity. 

Religious anthropologists would note that costly signals generate credibility. Walking long distances, living simply, renouncing comfort are high-cost behaviors.

Even secular observers respect visible sacrifice. 

Many Americans and Westerners generally might be  exhausted by ideological warfare; a polarized culture and combative attitudes. 

The monks offered:

• Spirituality without aggression

• Conviction without outrage

• Identity without hostility. 

They embodied transcendence without demanding that others take sides. 

Others might point to the power of ritual and aesthetics:

• Robes

• Chanting

• Silence

• Repetition

• Rhythm.

Modern life has stripped away most shared ritual outside of sports and entertainment. When people encounter sacred ritual in public space, it disrupts routine perception. It feels “set apart.”

Even nonreligious observers often experience:

• Calm

• Curiosity

• A sense of gravity

Religious scholars would say this is an encounter with the sacred breaking into ordinary space.

Monks project seriousness without heaviness. 

If the monks were male, some observers might note something subtle: they represented disciplined, gentle, self-controlled masculinity.

In a cultural moment confused about male identity, visible restraint combined with purpose is compelling. It signals strength under control, not dominance.

Emotion spreads socially. A group that radiates calm, smiles gently, and moves slowly literally lowers ambient anxiety.

The phrase “Walk for Peace” itself meets a primal desire. Peace is universally valued, even if defined differently.

Here’s a harder truth: people often project onto monks what they wish were true about themselves.

• Greater discipline

• Greater faith

• Greater simplicity

• Greater inner stillness

From a cultural perspective, the warm reception wasn’t random. It reflects:

• Spiritual hunger

• Fatigue with ideological conflict

• Desire for authenticity

• Attraction to embodied sacrifice

• Longing for visible goodness

When a group appears to live what others only talk about, people respond.

If you zoom out, the reception says at least as much about the culture as it does about the monks.

A society doesn’t warmly embrace ascetics unless it senses it’s missing something.

Which Language Model Do You Prefer?

Our choices of “favored” language models will probably remain somewhat idiosyncratic for a while, until some winnowing of market leaders occurs and a stable structure emerges. 

Most casual users will probably simply rely on ChatGPT and likely have no way to evaluate nuances of different engines. Others might have some familiarity with a few different models, but have difficulty explaining their impressions of the differences between models.

Also, models can change over time. Very early on, I found using Gemini frustrating. Though it was among the best for importing results to Google productivity apps, the other models have gradually gotten easier to use, in that regard. But ease of use was not the key performance indicator, at least for me. 

Since I do lots of forecasting-type work, Gemini’s earlier versions were often frustrating for refusal to produce such content. That does not seem to be the case for the latest models, so I assume there were short-term guardrails put into place for essentially-regulatory or other business reasons (such as avoiding the embarrassment of nonsensical, libelous or dangerous answers). 

For casual, everyday uses, though, I increasingly rely on Google search, sometimes with its AI Mode enabled, but often not even bothering to do so, as the results will include such results anyhow. 

So much for language models “killing search.”

As Gemini’s performance has improved, I find it use it more, and ChatGPT less. Grok does seem to provide more punchy, interesting commentary, but Perplexity or Claude seem better when I need to document sources. 

As a non-coder, I never can evaluate that use case. 

But here’s another take on the strengths of various models. 

source: Special Situations Research, Seeking Alpha, Bret Jensen

And market share varies when looking at enterprise or consumer adoption. Anthropic (Claude) has become a leader in enterprise model spending, with an estimated 40 percent market share as of late 2025.

ChatGPT remains dominant in the consumer space, with 74 percent of the consumer LLM market as of May 2025, but declining into the 60-percent range by early 2026.

Google's Gemini is gaining share in both consumer and enterprise market segments.

I'm not sure what is happening with open source (Meta Llama, Mistral, Chinese model share), but should be growing among enterprises seeking to avoid vendor lock-in and manage data privacy, or for academic users. 

Walk for Peace Reaches Washington, D.C.

 The #WalkforPeace monks have reached Washington, D.C. 

Moving Towards Generative User Interface

There’s a reason enterprise software has taken a beating in financial markets recently: nobody is sure how much value language models are going to destroy.

We are moving toward Generative UI, where the interface doesn't exist until you ask for it. If you need a specific chart, the LLM generates that specific chart in the chat window, for example. 

There are going to be lots of business model changes for enterprise and consumer software. 

Once the task is done, the interface disappears. This "ephemeral" UI is far more efficient than static dashboards, posing a direct threat to any software whose main value is "organizing data into screens."

Instead of static UI components, Generative UI introduces self-evolving interfaces that dynamically respond to user needs, much like how Generative AI models produce text, images, or code on demand, generating the application’s interface on the fly based on user intent.

By 2026, this technology is shifting the power dynamic from software vendors (who dictate workflows) to users.

Industry	Traditional Barrier	GenUI Disruption
Customer Relationship Management	Manual data management & "Tab Fatigue."	Outcome-based workspaces that appear only when needed.
Enterprise Resource Planning	Extreme complexity & high training costs.	Natural language translation of business data into simple "Action Cards."
Creative	Technical skill & "Steep Learning Curves."	Intent-driven canvases where the AI handles technical execution.

In a traditional CRM, sales reps spend up to 70 percent of their time navigating tabs, logging calls, and updating pipeline stages. GenUI replaces the static "account page" with an ephemeral workspace: just ask a question about a customer account. 

When a sales manager asks "which deals are at risk due to lack of executive engagement," GenUI doesn't just list them; it builds a temporary interface showing a side-by-side comparison of email sentiment, a "ghost" organizational chart of the client, and a pre-drafted calendar invite for a "check-in" meeting.

The concept of "searching for a record" disappears, as “the UI is the search.”

You talk to the CRM, and the specific fields you need to edit materialize in front of you, then vanish when the task is done.

ERPs have been difficult to navigate. GenUI democratizes the ERP by acting as a translator between complex business logic and human intent.

A procurement officer sees a news alert about a port strike. Instead of digging through Oracle's supply chain module, they ask the GenUI to "visualize the impact on our Q3 inventory." 

The system instantly renders a custom map and a "what-if" slider tool that lets the user simulate different shipping routes—functionality that might have taken a developer weeks to build as a permanent feature.

For reconciliation or expense audits, instead of a spreadsheet of 10,000 rows, the interface generates a "review card" for the five most suspicious transactions, with integrated buttons to "Approve," "Flag," or "Ask Employee for Receipt."

Creative software such as Adobe can take years to master. In web or UI design (Adobe XD or Figma), a designer can say, "Create a high-fidelity checkout page for a luxury watch brand." The GenUI generates editable layers, buttons, and cascading style sheets. 

Industry	Traditional Barrier	GenUI Disruption
CRM	Manual data management & "Tab Fatigue."	Outcome-based workspaces that appear only when needed.
ERP	Extreme complexity & high training costs.	Natural language translation of business data into simple "Action Cards."
Creative	Technical skill & "Steep Learning Curves."	Intent-driven canvases where the AI handles technical execution.

But an AI interface can potentially deliver all these capabilities through natural language, collapsing the feature hierarchy that supported tiered pricing models. 

On the other hand, there are cost issues distinct from traditional software as a service, where serving additional users costs almost nothing.

A company providing AI-powered customer service might pay $0.50-$2.00 per complex interaction in application programming interface costs alone. This fundamentally changes unit economics, as costs scale with usage intensity, not just user count. 

When software products use similar underlying models (Claude, GPT-4 and others), differentiation also becomes an issue. Why pay for ten different AI-powered tools when they're all essentially wrappers around the same language model?

So revenue is challenged while costs grow. 

Software Category	Traditional Revenue Model	AI-Induced Challenge	Potential Adaptation
CRM Systems	Per-seat licensing plus tier-based features (Basic/Pro/Enterprise)	AI can deliver "Enterprise" insights to Basic users; computational costs scale with data analysis	Usage-based pricing on AI features; charge for proprietary data connections and workflows
Project Management	Tiered subscriptions based on team size and features	Natural language interface collapses feature differentiation between tiers	Shift to charging for outcomes (projects delivered, efficiency gains) rather than features
Legal Research	Flat subscription or per-search fees	General LLMs can perform basic legal research; commoditizes core product	Focus on verified, citation-quality results; charge premium for liability/accuracy guarantees
Business Intelligence	Per-user licenses and data volume tiers	AI democratizes analytics; hard to charge more for "advanced" users who just ask better questions	Charge for data integration complexity, governance features, and certified insights rather than analysis capability
Customer Support	Per-agent seat licenses	AI reduces headcount needs (fewer seats sold); usage costs rise with ticket volume	Shift to per-resolution or per-customer pricing; charge for AI training on company data
Writing Tools	Monthly subscription ($10-30)	Directly competes with ChatGPT/Claude at $20/month with broader capabilities	Specialize in specific domains (academic, technical); integrate tightly with existing workflows
Code Editors/IDEs	Freemium or one-time purchase	AI coding assistants add significant per-user compute costs	Usage-based pricing on AI features while keeping base editor affordable
Design Software	Perpetual license or subscription	AI generation features expensive to operate; threatens margins on traditional tools	Separate pricing for generative AI features; charge for commercial usage rights
HR/Recruiting	Per-job-posting or per-hire fees	AI can screen resumes and match candidates, but at compute cost per evaluation	Charge for quality of matches and time-to-hire improvement rather than volume
Email, Productivity	Bundled suite pricing	AI features (smart compose, summarization) add costs that vary dramatically by user	Tiered AI quotas; charge power users more for intensive AI feature usage

Enterprise customers may be more tolerant of usage-based pricing since they're accustomed to paying for value delivered. 

But consumer products face harsher constraints. Users expect fixed, predictable monthly fees and react negatively to usage limits.

The fundamental question remains: as AI capabilities become more uniform and accessible, how do software companies justify premium pricing? The answer likely involves some combination of specialized data, deep workflow integration, reliability guarantees, and human expertise.

But all that introduces new levels of uncertainty into the value and valuation of enterprise software companies.